The Ground Game - Week 6

The “Ground Game”, Political Campaigns’ strategy to mobilize and persuade voters is critically important when trying to win elections. While persuading voters is becoming increasingly difficult as the country becomes more partisan, both political parties are realizing that success is predicated on turnout. There have been many large scale “get-out-the-vote” campaigns in recent years. For example, in 2018 during the Gubnatorial race, Stacey Abrams’s Campaign helped register hundreds of thousands of voters

From the plot below, we can see that congressional districts, even in the same state, have very different levels of turnout. For example, in Texas, we can see that rural regions have lower turnout (corresponding to purple and blue shades). Urban areas have higher turnout, indicated by brighter shades of red. At first glance, this map seems to confirm an assumption both the Democrats and Republicans operate under: higher turnout helps the Democrats. This is because in rural states that are more conservative, turnout appears to be lower than urban areas. However, this concept may be misleading. For example, even though a rural, conservative district may have low turnout, there may not necessarily be liberal people who do not vote. Instead, it is possible that the district is politically homogenous, and higher turnout would lead to more conservatives voting. We can see this from the plot of contested races in 2018 compared to turnout. There is only a small linear relationship between turnout and Democratic vote share, meaning increased turnout only has a small positive relationship with Democratic success.

Building a Model - While turnout may not be useful in a linear model, we can use recent polls as a predictor and then simulate turnout for each party. In this method, we make a generalized linear model for each district, with every recent poll as an observation. Doing 10,000 simulations for each district, using the poll results as the probability of someone voting for a Democrat or Republican, we can get an average margin of victory in each district. From the histograms below, we can see that there are a wide range of outcomes for each seat that we have data for (32 districts in total). Unfortunately, when observing these results, we can see that this model may be problematic. For example, in many instances, neither party gets close to 50% of the vote. Obviously, this does not make sense because in most cases, there is only a Democrat and Republican candidate. While we will eventually predict all the house seats, in our district-level map, it probably makes sense to ignore these results for now. While simulation may be a good idea as we approach our final prediction, turnout is clearly not a useful predictor.

state st_cd_fips pred_dem pred_rep mean_democrat_win_margin
California 0622 0.2220354 0.2906073 -17.1717172
Florida 1213 0.3686113 0.3263571 5.7665260
Florida 1227 0.3166663 0.2684526 13.9372822
Illinois 1713 0.2593444 0.3457982 -15.7522124
Iowa 1901 0.3046030 0.2418445 10.6761566
Iowa 1902 0.3064553 0.2736335 2.3648649
Iowa 1903 0.3158108 0.2550524 8.7033748
Kansas 2003 0.3521102 0.3032105 6.6465257
Maine 2301 0.4947502 0.2556307 31.8681319
Maine 2302 0.3014435 0.3196930 4.1269841
Michigan 2603 0.3459113 0.3894398 -11.3233288
Michigan 2608 0.3158895 0.2919566 12.5208681
Minnesota 2701 0.3234932 0.3771824 -11.6618076
Minnesota 2702 0.3653585 0.3655082 2.4251070
Minnesota 2703 0.3995444 0.3174953 12.8939828
Nebraska 3102 0.3017581 0.5467812 -26.5550239
Nevada 3201 0.2462418 0.1148674 34.7181009
Nevada 3202 0.2363129 0.3294454 -9.8445596
Nevada 3203 0.2635670 0.2175362 12.2137405
New Jersey 3403 0.2823976 0.2747623 4.1591320
New Jersey 3407 0.3267295 0.3243389 -4.3478261
New Mexico 3502 0.2285342 0.1719934 16.8269231
New York 3611 0.2166927 0.1432826 27.9329609
New York 3619 0.2680315 0.2408428 6.5868263
North Carolina 3711 0.3166000 0.4075260 -10.4046243
North Carolina 3713 0.2307081 0.2610819 -1.6460905
Ohio 3901 0.2767639 0.2901091 -0.5545287
Pennsylvania 4201 0.3139580 0.3282828 1.2307692
Pennsylvania 4208 0.3152016 0.1779728 21.9616205
Pennsylvania 4217 0.4006544 0.3826833 2.2443890
Virginia 5102 0.2633871 0.2366750 2.4691358
Washington 5308 0.3240912 0.2942258 0.0000000

Looking Ahead

Going forward, I think pooling all the past congressional races (for which we have data) makes the most sense for predicting the upcoming midterms. While grouping the models by district makes sense in theory, I have not added any data that helps describe the characteristics of a district. For example, we have yet to incorporate levels of education, or how rural or urban the district is. Instead, like in this exercise, I made a model for each district, but each one had very few observations, leading to results that did not make sense.